A Spatial Non-Cooperative Target Acquisition Method Based on Deep Reinforcement Learning

Abstract

The invention discloses a space non-cooperative target capture method based on deep reinforcement learning, which is completed in two steps and realizes interaction. The first step is to use 3D visualization software to build a 3D visualization environment for the service aircraft and the target aircraft. The input of the visualization environment is the control force and control torque of the service aircraft, and the output is the state of the service aircraft and the target aircraft. The second step is to construct a convolutional neural network model, and conduct intelligent autonomous space non-cooperative target acquisition training for the service aircraft in a 3D visualization environment. The convolutional neural network model takes the state of the service aircraft and the target aircraft as input, and uses its weight parameters to output the control force and control torque required to control the service aircraft and send them into the visualization environment. The states of the two aircraft continue to be input into the neural network for continuous deep reinforcement training. Through the continuous interaction between the visual environment and the neural network, the feedback results are captured and output correctly.

Description

technical field [0001] The invention belongs to the field of aerospace technology, and in particular relates to a space non-cooperative target acquisition method based on deep reinforcement learning. Background technique [0002] Non-cooperative targets refer to spacecraft that are not designed for docking or capture, such as satellites and space debris that are not equipped with cooperative components of one’s own side, as well as spacecraft of the other party. They do not communicate at the information level and do not cooperate in maneuvering behavior. challenge. Many space military missions, such as destroying enemy space vehicles and assisting satellites that have not successfully entered the predetermined orbit, etc., need to complete the on-orbit capture of non-cooperative targets first. [0003] Judging from the current development situation, the capture technology for space cooperative targets has been relatively mature and has been successfully applied in on-orbit...

AI Technical Summary

Problems solved by technology

However, these methods have certain limitations: either the target model is known; or the image processing on the ground is required, and then the data is uploaded to the star, there is a certain time delay, and the reliability is affected; or it can only target a certain type of target, have limitations

Images